An exploratory analysis of phrases in text retrieval
Jeremy Pickens, W. Bruce Croft · 2000
Phrases are used in both commercial and experimental search engines. Despite the large amount of work in the area results remain mixed. It is still not clear whether phrases can be used to improve retrieval effectiveness. In this paper, we examine phrases and their properties independently of any specific retrieval approach. We explore phrase usage in text corpora and relevance patterns related to phrase usage. The result is not only a better understanding of phrases, but a better method by which phrases and phrase techniques may be evaluated. With this method we can directly determine the value of various phrase formulations for information retrieval. 1 Introduction Phrases in information retrieval have been used to aid effectiveness. They have gone by many names: multiword features, phrase terms, nominal compounds and so on. Despite their many implementations, retrieval experiments with phrases are often inconclusive. In some experiments phrases yield small improvements in re...